Each time a CPU acquire a mutex, just after the memory
barrier, it can check if other threads published new
pages, and update its own view. Thus, you get the page
trap only for programs that are not well synchronized.
Also, there is no need to publish per object. It must
publish when releasing a mutex. At all other cases, it
can delay the publishing.
Doron Rajwan
--- Eliot Moss <moss@cs.umass.edu> wrote:
> Sheesh, I wouldn't want to pay the thousands to
> millions of instructions
> for handling a page trap on publishing each pointer.
> That's a huge
> per-object overhead ..... Eliot Moss
> -------------------------------
> JavaMemoryModel mailing list -
http://www.cs.umd.edu/~pugh/java/memoryModel
=====
Doron Rajwan, mailto:doron@rajwan.org
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:51 EDT